Frontier LLMs pass fewer than 58% of systematically varied safety-fact scenarios, revealing weak generalization of critical safety knowledge to naive user queries.
Agentharm: A benchmark for measuring harmfulness of llm agents, 2024
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
SAGE-Eval: Evaluating LLMs for Systematic Generalizations of Safety Facts
Frontier LLMs pass fewer than 58% of systematically varied safety-fact scenarios, revealing weak generalization of critical safety knowledge to naive user queries.